TW201429227A - Noise estimation apparatus and method thereof - Google Patents

Abstract

A noise estimation apparatus and method thereof for calculating a noise estimation value of an image. The noise estimation apparatus comprises a calculating unit, a variance calculating unit, a distribution curve generating unit and a noise estimation unit. The calculating unit generates a pixel distribution according to a plurality of pixel data of an i<SP>th</SP> block of the image and a plurality of pixel data of the i<SP>th</SP> block of a previous image. The variance calculating unit combines the pixel data of an i<SP>th</SP> block of the image and the pixel data of an i<SP>th</SP> block of the previous image to correspondingly generate a variance value. The distribution curve generating unit generates a distribution curve according to the variance value, and compares the pixel distribution with the distribution curve to correspondingly generate a weighting value. The noise estimation unit outputs the noise estimation value according to the weighting value and the variance value corresponding to each of the blocks of the image.

Description

Noise estimation device and method thereof

The present invention relates to a noise estimation apparatus and method thereof, and more particularly to a noise estimation apparatus for calculating a noise estimation value of an image and a method thereof.

In recent years, due to the rapid development of multimedia applications, the requirements for image quality have also increased. However, multimedia images are often interfered with by noise. These noises not only degrade the quality of the image, but also reduce the sharpness and sharpness of the image, and the outline of the object becomes blurred. In order to eliminate image noise, noise estimation technology is commonly used in image processing systems as a benchmark for noise reduction processing.

However, when performing noise estimation, the details or dynamic images in the image are often misjudged as noise in the image. If the result is judged and then the noise reduction processing is performed, the noise will be reduced. The processed image is instead reduced in image quality.

Therefore, how to provide a noise estimation technology that can effectively judge the degree of noise interference of image content to avoid image details or motion images being misjudged as noise is one of the topics of the industry.

The invention relates to a noise estimation technology, which can effectively judge the degree of interference of image content by noise.

According to an aspect of the invention, a noise estimation apparatus is provided. The noise estimation device is configured to calculate a noise estimation value of an image, the image has M blocks, the M blocks may overlap or not overlap, and each block has a plurality of pixel data, and the blocks The pixel data includes a target pixel data, and the M system is a positive integer greater than 1. The noise estimation device includes a distribution calculation unit, a mutation calculation unit, a distribution curve generation module, and a noise estimation unit. The distribution calculation unit is configured to calculate pixels corresponding to the plurality of pixel values according to the pixel data of the i-th block of the image and the plurality of previous pixel data of the i-th block of a previous image. Number to produce a pixel distribution, i is a positive integer between 1 and M. The mutation calculation unit is configured to combine the pixel data of the i-th block of the image and the previous pixel data of the i-th block of the previous image, and correspondingly generate a variance. The distribution curve generation module generates a distribution curve based on the target pixel data of the i-th block of the image, and compares the pixel distribution and the distribution curve to correspondingly generate a weight value. The noise estimation unit is configured to output the noise estimation value according to the weight value and the variance corresponding to each block of the image.

According to another aspect of the present invention, a noise estimation method is provided for use in a noise estimation device for calculating a noise estimation value of an image having M blocks, The M blocks may overlap or not overlap, and each block has a plurality of pixel data, the pixel data includes a target pixel data, and the M system is a positive integer greater than 1. The noise estimation method includes: Calculating the pixel numbers of the i-th block of the image and the plurality of previous pixel data of the i-th block of the previous image, and calculating the number of pixels corresponding to the plurality of pixel values to generate an image a distribution of i, a positive integer between 1 and M; combining the pixel data of the i-th block of the image with the previous pixel data of the i-th block of the previous image, and correspondingly Generating a variogram; generating a distribution curve based on the target pixel data of the i-th block of the image, and comparing the pixel distribution and the distribution curve to correspondingly generate a weight value; And outputting the noise estimation value according to the weight value and the variance corresponding to each block of the image.

In order to provide a better understanding of the above and other aspects of the present invention, the following detailed description of the embodiments and the accompanying drawings

Please refer to FIG. 1. FIG. 1 is a block diagram of a noise estimation apparatus 100 according to an embodiment of the present invention. The noise estimation device 100 is configured to calculate the noise estimation value σ _{avg of the} image. This image has M blocks that may or may not overlap. Each block has a plurality of pixel data, and the pixel data includes a target pixel data, and M is a positive integer greater than 1. The plurality of pieces of pixel data described above are, for example, corresponding to a plurality of pixel points on the display panel.

For ease of understanding, please also refer to Figures 2A and 2B. 2A and 2B are schematic diagrams showing that M blocks of an image can be overlapped and not overlapped, respectively. In Fig. 2A, each row in the image Ft has a pixel column, and each small square in each row represents one pixel. Each pixel system corresponds to a block, and the target pixel data is a central pixel data of the corresponding block. Further, if the size of the block in the image is a pixel column of 1×K, and K is an odd number greater than zero, when the target pixel data is the i-th pixel data P _i in the image, the pixel data corresponds to the pixel data. the i-th image blocks P _i of Ft (i) comprises the pixel data _{P i- (K-1) /} 2, P i- (K-1) / 2 + 1, ... P i ... , P _i+(K-1)/2-1 , P _i+(K-1)/2 . When the target pixel data is the i+1th pixel data P _i +1 in the image, the i+1th block Ft(i+1) in the image corresponding to the pixel data Pi+1 includes the pixel data. P _i+1-(K-1)/2 , P _{i+1-(K-1)/2+1} , ...P _i+1 ..., P _{i+1+(K-1)/ 2-1} , P _i+1+(K-1)/2 , and so on. On the other hand, in FIG. 2B, the size of the block in the image is, for example, a pixel matrix of K×K, and the image Ft′ is equally divided into a plurality of blocks that do not overlap each other, and the target pixel data of each block is, for example, The central pixel data of the block. For example, the target pixel data of the block Ft'(i) is, for example, the pixel data P _i , and the target pixel data of the block Ft'(i+1) is, for example, the pixel data P _i+K . However, the present invention is not limited thereto, and the size of the above block and the position of the target pixel data relative to its corresponding block may be adjusted according to different applications.

Referring to FIG. 1 again, the noise estimation apparatus 100 includes a distribution calculation unit 102, a variation calculation unit 104, a distribution curve generation module 106, and a noise estimation unit 108. The distribution calculation unit 102 is configured to use the plurality of pieces of pixel data of the i-th block Ft(i) of the image Ft and the plurality of previous pixel data of the i-th block Ft-1(i) of the previous image Ft-1. The number of pixels corresponding to the plurality of pixel values is calculated to generate a pixel distribution h(i), and i is a positive integer between 1 and M. The i-th block Ft(i) of the image Ft and the i-th block Ft-1(i) of the previous image Ft-1 correspond, for example, to the same display position in the picture, and in time ordering, previously The image Ft-1 represents an image displayed earlier than the image Ft. Further, the above pixel value is, for example, a gray scale level.

The mutation calculation unit 104 is configured to combine the plurality of pieces of pixel data of the i-th block Ft(i) of the image Ft and the plurality of previous pixel data of the i-th block Ft-1(i) of the previous image Ft-1, And correspondingly produces a variance σ(i). The above-mentioned variation number σ(i) represents, for example, a statistical variation.

The distribution curve generation module 106 generates a distribution curve G(i) according to the variation number σ(i) based on the target pixel data of the i-th block Ft(i) of the image Ft, and compares the pixel distribution h(i). And a distribution curve G(i) to correspondingly generate a weight value W(i).

Finally, the noise estimation unit 108 is configured to use the weight values W(1)~W(M) and the variance σ(1)~σ corresponding to each block Ft(1)~Ft(M) of the Ft image. (M), output noise estimation value σ _avg .

Referring to FIG. 3, an example of a block diagram of the above-described distribution calculation unit 102 is shown. The distribution calculation unit 102 includes a first calculation unit 302, a second calculation unit 304, and a synthesis unit 306. The first calculating unit 302 is configured to calculate the number of pixels corresponding to each of the plurality of pixel data of the i-th block of the image Ft to output a first distribution h1(i). The second calculating unit 304 is configured to calculate the number of pixels corresponding to each of the plurality of previous pixel data of the i-th block of the previous image Ft-1 to output a second distribution h2(i). The synthesizing unit 306 is configured to combine the first distribution h1(i) and the second distribution h2(i) to generate a pixel distribution h(i). The first distribution h1(i), the second distribution h2(i), and the pixel distribution h(i) described above are, for example, in the form of a histogram.

Please refer to both Figure 4A and Figure 4B. FIG. 4A is a schematic diagram showing an example of the first distribution h1(i), and FIG. 4B is a schematic diagram showing an example of the second distribution h2(i). In Figure 4A and Figure 4B, the coordinates of the coordinates The horizontal axis represents the grayscale value, and the range is, for example, 0 to 255; the vertical axis of the coordinates represents the number of pixels. Each straight square bar represents the number of pixels corresponding to a gray scale value.

The combination unit 306 combines the first distribution h1(i) and the second distribution h2(i), for example, to superposition the two distributions. For example, the number of pixels corresponding to each pixel value in the first distribution h1(i) is separately added to the number of pixels corresponding to each pixel value in the second distribution h2(i) to generate a corresponding difference. The pixel distribution h(i) of the number of pixels of the pixel value. As shown in FIG. 4C, the pixel distribution h(i) generated by superimposing the first distribution h1(i) in FIG. 4A and the second distribution h2(i) in FIG. 4B is shown. A schematic diagram of an example.

However, the present invention is not limited to the above-described example, and the distribution calculation unit 102 may also process a plurality of sets of pixel data corresponding to the i-th block of the pictures Ft~Ft-k respectively corresponding to different time points, to correspondingly A pixel distribution h'(i) is generated, where k is a positive integer. The introduction of multiple pictures Ft~Ft-k corresponding to different time points respectively can provide more sampling points in image processing.

Referring to FIG. 5, an example of a block diagram of the variation calculation unit 104 is shown. The mutation calculation unit 104 includes an operation unit 502 and a variation number acquisition unit 504. The computing unit 502 is configured to combine the plurality of pieces of pixel data of the i-th block of the image Ft with the plurality of pieces of previous pixel data of the i-th block of the previous image Ft-1 to correspondingly output a synthesized pixel data S(i).

Please refer to FIG. 6 , which illustrates a schematic diagram of an example of the synthesized pixel data S(i). In FIG. 6, the plurality of pieces of pixel data Ft(i) of the i-th block of the image Ft are connected to the plurality of previous pixel data Ft-1(i) of the i-th block of the previous image Ft-1, To combine into a single synthetic pixel data S(i). On The exemplification is not intended to limit the present invention, and the operation unit 502 may also generate other pieces of pixel data Ft(i) including the i-th block of the image Ft and the previous image Ft-1 through other combinations and sorting methods. The synthesized pixel data S(i) of the plurality of previous pixel data Ft-1(i) of the i-th block.

Referring again to FIG. 5, the variance obtaining unit 504 is configured to generate the variation σ(i) based on the synthesized pixel data S(i). Since the manner in which the variation σ(i) is generated can be obtained, for example, by a general image processing software, it will not be described here. However, if the distribution calculation unit 102 processes the plurality of sets of pixel data corresponding to the i-th block of the pictures Ft~Ft-k respectively corresponding to different time points according to the above, to correspondingly generate the pixel distribution h'(i), Then, the mutation calculation unit 104 correspondingly calculates a plurality of pixel data of the i-th block of the pictures Ft~Ft-k at different time points to generate a variation σ'(i). In short, the embodiment of the present invention is not limited to the noise estimation based on the image Ft and the previous image Ft-1, and may also be used as the noise according to multiple groups of pictures Ft~Ft-k corresponding to different time points. Estimate.

Please refer to FIG. 7 , which illustrates an example of a block diagram of the distribution curve generation module 106 . The distribution curve generation module 106 includes a distribution curve generation unit 702, a difference calculation unit 704, and a weight calculation unit 706. The distribution curve generation unit 702 performs a lookup table based on the target pixel data of the i-th block of the image Ft based on the variation number σ(i) to output a distribution curve G(i).

The above-described distribution curve G(i) is, for example, a Gaussian distribution. The reason why the Gaussian distribution is selected as an example is that the applicant found in the research that among the general TV image signals, the pixel data interfered by the noise is often presented in a Gaussian distribution; relatively, the content of the image The detail portion or the pixel distribution corresponding to the motion picture is usually quite different from the Gaussian distribution. Therefore, if the pixel distribution in the image content is more similar to the Gaussian distribution after comparison, it indicates that the image content is more likely to be interfered by the noise. However, the present invention is not limited thereto, and in another embodiment, the distribution curve G(i) may also be, for example, a Poisson distribution. In summary, the curve of the distribution can correspond to the noise model in the image signal without departing from the spirit of the distribution curve G(i) in the embodiment of the present invention.

In addition, if the distribution curve G(i) is a Gaussian distribution, the Gaussian distribution has a fixed coverage area under the distribution curve. Through this characteristic, the array can be pre-established to correspond to different variograms σ(i). The Gaussian distribution model is included in the distribution curve generation unit 702 for subsequent lookup tables.

Please also refer to Fig. 8, which is a schematic diagram showing an example of the distribution curve G(i). In Fig. 8, the horizontal axis of the coordinates represents a pixel value, and the pixel value is, for example, a grayscale value, the range of which is, for example, 0 to 255, and the vertical axis represents the number of pixels. A curve 802 represents a distribution curve G(i) obtained by looking up the table based on the variation σ(i) with the pixel value Pv corresponding to the target pixel data as a reference. In this embodiment, since the pre-established distribution model (for example, a Gaussian distribution model) among the distribution curve generation units 702 is based on the coordinate origin (for example, the mean value is zero), in order to present the reference pixel data as a reference. The distribution model, the distribution curve generation module 106 further sets the pixel value Pv corresponding to the target pixel data as the reference of the distribution model to be output as the distribution curve G(i).

Referring again to FIG. 7, the difference calculation unit 704 is configured to compare the distribution curve G(i) and the pixel distribution h(i) to output a difference value D(i). The difference value D(i) can be obtained, for example, by:

Wherein, the parameter k represents a pixel value, and if the pixel value represents a grayscale value, the range of the parameter k is, for example, 0 to 255; the parameter _Sk represents the number of pixels of the distribution curve G(i) at the pixel value k minus the pixel distribution h (i) The number of pixels at the pixel value k. Of course, the present invention is not limited thereto, and is a combination of a linear or non-linear combination of the parameters S _k or a value obtained by quantizing the difference between the distribution curve G(i) and the pixel distribution h(i), Both can be used as the difference value D(i) of the embodiment of the present invention.

The weight calculation unit 706 is configured to generate the weight value W(i) correspondingly according to the difference value D(i). Please refer to FIG. 9A, which is a schematic diagram showing an example of the correspondence between the weight value W(i) and the difference value D(i). As can be seen from the change of the curve 902, when the magnitude of the difference value D(i) is between the first threshold TH1 and the second threshold TH2, and the second threshold TH2 is greater than the first threshold TH1, the difference is When the value D(i) is larger, the corresponding weight value W(i) is smaller. In this embodiment, the weight value W(i) is between the weight upper limit value WH and the weight lower limit value WL. In the curve 902, between the first threshold TH1 and the second threshold TH2. The difference value D(i) is obtained by linear interpolation of the point coordinates (TH1, WH) and the point coordinates (TH2, WL). The weight upper limit value WH and the weight lower limit value WL are, for example, values of 0 to 1, respectively, and the weight upper limit value WH is greater than the weight lower limit value WL. The significance of the curve 902 is that the weight value W(i) corresponds, for example, to the probability of occurrence of noise. When the difference value D(i) is larger, the difference between the pixel distribution h(i) and the distribution curve G(i) is larger. , It also means that the less the pixel distribution h(i) is similar to the distribution curve of the noise (for example, the Gaussian curve), the lower the probability of occurrence of noise, so the weight value W(i) is correspondingly smaller. However, the present invention is not limited to the above examples, that is, the correspondence relationship between the weight value (i) and the difference value D(i) is not limited to linear interpolation between two points, as long as the weight value W(i) The weight of the noise can be reflected according to the difference value D(i), which is within the scope of the spirit of the present invention.

The first threshold value TH1 and the second threshold value TH2 described above are generated, for example, by a dynamic threshold according to the variation number σ(i). For the sake of clarity, please refer to FIG. 9B and FIG. 9C simultaneously, which respectively show a schematic diagram of a correspondence relationship between the first threshold TH1 and the second threshold TH2 and the variation σ(i), respectively. In Fig. 9B, it can be seen from the change of the curve 904 that when the variation σ(i) varies between the lower threshold THa and the upper threshold THb, the first threshold TH1 is between the first lower limit TH1α and the first Between an upper limit value TH1β, the value of the first threshold TH1 can be formed by linear interpolation of points (THa, TH1α) and points (THb, TH1β). Similarly, in FIG. 9C, it can be seen from the change of the curve 906 that when the variation σ(i) varies between the lower threshold THa and the upper threshold THb, the second threshold TH2 is below the second lower limit. Between the TH2α and the second upper limit value TH2β, the value of the second threshold TH2 is formed by linear interpolation of the points (THa, TH2α) and the points (THb, TH2β). The first lower limit value TH1α, the first upper limit value TH1β, the second lower limit value TH2α, the second upper limit value TH2β, the upper limit value THb, and the lower limit value THa may be set according to different application conditions, but the first Both the lower limit value TH2α and the second upper limit value TH2β must be greater than the first lower limit value TH1α and the first upper limit value TH1β, respectively. The advantage of generating the first threshold TH1 and the second threshold TH2 in a dynamic threshold is that when the resulting variation σ(i) is too large At the same time, the weight calculation unit 706 can also determine the appropriate weight value W(i) according to the difference value D(i) to effectively determine the degree of noise interference of the video content.

Referring to FIG. 10, a block diagram of an example of the noise estimation unit 108 is shown. The noise estimation unit 108 includes a noise estimator 1002. The noise estimator 1002 is configured to use the weight values W(1)~W(M) and the variance σ(1)~σ(M) corresponding to each of the blocks Ft(1)~Ft(M) of the image Ft. A weighted averaging process is performed to output a noise estimation value σ _avg . For example, the noise estimation value σ _avg can be obtained, for example, by:

As can be seen from the above equation, the noise estimation value σ _avg indicates the degree to which the entire image Ft is disturbed by noise. When the noise estimation value σ _avg is larger, it indicates that the image Ft is disturbed by noise, and conversely, the smaller the noise estimation value σ _{avg , the} lower the degree of noise interference of the image Ft. However, the present invention is not limited thereto, and any result obtained by combining the weight value W(i) and the variation number σ(i) can be used as a reference value for evaluating image noise in the present invention.

In addition, the noise estimation unit 108 may further include a protection unit 1004. The protection unit 1004 is used for summing the weight values W(1)~W(M) of all the blocks Ft(1)~Ft(M) (ie When the value is lower than a threshold value T, the noise estimation value σ _avg outputted by the noise estimator 1002 is divided by a low-decrease DN to be output as the noise estimation value σ _{avg '} . Since the sum of the weight values W(1)~W(M) of all the blocks Ft(1)~Ft(M) is too low, the corresponding noise estimation value σ _avg is less reliable ( Unreliable, in this case, the protection unit 1004 divides the noise estimation value σ _avg by the low depreciation DN to appropriately adjust the noise estimation value σ _avg as an output.

Please refer to FIG. 11 , which is a schematic diagram showing an example of the correspondence between the sum of the weight values W(1) to W(M) of all the blocks Ft(1) to Ft(M) and the low depreciation DN. As can be seen from the curve 1102, when the sum of the weight values W(1) to W(M) of all the blocks Ft(1) to Ft(M) is lower than the threshold value T, the lower the value of the DN is larger. The range of the low depreciation DN is, for example, between the low decrement upper limit value DNH and the low decrement lower limit value DNL. The value of the low-reduction upper limit value DNH is, for example, 128, and the value of the low-subtraction lower limit value DNL is, for example, 1.

In this embodiment, a noise estimation method is further proposed, and the noise estimation method is used in the noise estimation apparatus 100 of the embodiment of the present invention. Please refer to FIG. 12, which is a flow chart of the noise estimation method of the embodiment. The method includes steps S1202, S1204, S1206, and S1208. First, in step S1202, according to the plurality of pixel data of the i-th block of the image Ft and the plurality of previous pixel data of the i-th block of the previous image Ft-1, the corresponding correspondences of the plurality of pixel values are respectively calculated. The number of pixels is used to produce a pixel distribution h(i), which is a positive integer between 1 and M.

Next, in step S1204, the plurality of pieces of pixel data of the i-th block of the image Ft and the plurality of previous pixel data of the i-th block of the previous image Ft-1 are combined, and a variation σ(i) is correspondingly generated. ).

Then, in step S1206, based on the target pixel data of the i-th block of the image Ft, a distribution curve is generated according to the variation σ(i). G(i), and compares the pixel distribution h(i) and the distribution curve G(i) to correspondingly generate a weight value W(i).

Finally, in step S1208, the weight values W(1)~W(M) and the variance numbers σ(1)~σ(M) corresponding to each of the blocks Ft(1)~Ft(M) of the image Ft are obtained. , output noise estimation value σ _avg .

The noise estimation device and method for calculating the noise estimation value of the image in the embodiment can effectively determine the degree of noise interference of the image content.

In conclusion, the present invention has been disclosed in the above embodiments, but it is not intended to limit the present invention. A person skilled in the art can make various changes and modifications without departing from the spirit and scope of the invention. Therefore, the scope of the invention is defined by the scope of the appended claims.

100‧‧‧ Noise Estimation Device

102‧‧‧Distribution calculation unit

104‧‧‧variation calculation unit

106‧‧‧Distribution curve generation module

108‧‧‧ Noise Estimation Unit

302‧‧‧First calculation unit

304‧‧‧Second calculation unit

306‧‧‧Synthesis unit

502‧‧‧ arithmetic unit

504‧‧‧variation acquisition unit

702‧‧‧Distribution curve generation unit

704‧‧‧ difference calculation unit

706‧‧‧ weight calculation unit

1002‧‧‧ Noise Estimator

1004‧‧‧protection unit

FIG. 1 is a block diagram of a noise estimation apparatus according to an embodiment of the present invention.

FIG. 2A is a schematic diagram showing that M blocks of an image can overlap.

FIG. 2B is a schematic diagram showing that the M blocks of the image do not overlap.

Figure 3 is a block diagram showing an example of a distributed calculation unit.

FIG. 4A is a schematic diagram showing an example of the first distribution.

FIG. 4B is a schematic diagram showing an example of the second distribution.

FIG. 4C is a schematic diagram showing an example of a pixel distribution.

Figure 5 is a block diagram showing an example of a variation calculation unit.

FIG. 6 is a schematic diagram showing an example of synthesized pixel data.

Figure 7 is a block diagram showing an example of a distribution curve generating module.

Figure 8 is a schematic diagram showing an example of a distribution curve.

FIG. 9A is a schematic diagram showing an example of the correspondence between the weight value and the difference value.

FIG. 9B is a schematic diagram showing an example of the correspondence between the first threshold value and the variation number.

FIG. 9C is a schematic diagram showing an example of the correspondence between the second threshold value and the variation number.

Figure 10 is a block diagram showing an example of a noise estimation unit.

FIG. 11 is a schematic diagram showing an example of the correspondence between the sum of the weight values of all the blocks and the low depreciation.

FIG. 12 is a flow chart showing the method for estimating noise in the embodiment.

100‧‧‧ Noise Estimation Device

102‧‧‧Distribution calculation unit

104‧‧‧variation calculation unit

106‧‧‧Distribution curve generation module

108‧‧‧ Noise Estimation Unit

Claims (20)

A noise estimation device for calculating a noise estimation value of an image, the image having M blocks, the M blocks may overlap or not overlap, and each block has a plurality of pixel data, The pixel data includes a target pixel data, and the M is a positive integer greater than 1. The noise estimation device includes: a distribution calculation unit configured to use the pixel data of the i-th block of the image, and a plurality of previous pixel data of the i-th block of a previous image, calculating a number of pixels corresponding to the plurality of pixel values to generate a pixel distribution, i is a positive integer from 1 to M; a variation a calculating unit, configured to combine the pixel data of the i-th block of the image and the previous pixel data of the i-th block of the previous image, and correspondingly generate a variance; a distribution curve is generated The module generates a distribution curve based on the target pixel data of the i-th block of the image, and compares the pixel distribution and the distribution curve to correspondingly generate a weight value; and Noise estimation list , Corresponding to the number and the weight value based on the variation of the image of each block, the output of the noise estimate. The noise estimation device of claim 1, wherein the distribution calculation unit comprises: a first calculation unit, configured to calculate corresponding to each of the pixel data of the i-th block of the image The number of pixels of the pixel values to output a first distribution; a second calculation unit for calculating the ith region of the previous image And the number of pixels of the previous pixel data of the block corresponding to each of the pixel values to output a second distribution; and a combining unit for combining the first distribution and the second distribution to generate the pixel distribution . The noise estimation device of claim 1, wherein the variation calculation unit comprises: an operation unit configured to combine the pixel data of the i-th block of the image with the previous image The previous pixel data of the i-th block is correspondingly outputted to a synthesized pixel data; and a variance obtaining unit is configured to generate the variation according to the synthesized pixel data. The noise estimation device of claim 1, wherein the distribution curve generation module comprises: a distribution curve generation unit, based on the target pixel data of the i-th block of the image, The variation is performed by looking up the table to output the distribution curve; a difference calculation unit for comparing the distribution curve and the pixel distribution to output a difference value; and a weight calculation unit for using the difference value to The weight value is generated correspondingly. The noise estimation device of claim 4, wherein the distribution curve is a Gaussian distribution. The noise estimation device of claim 4, wherein when the magnitude of the difference is between a first threshold and a second threshold, the second threshold is greater than the first valve Value, when the difference value is larger, the corresponding The smaller the weight value. The noise estimation device of claim 6, wherein the first threshold value and the second threshold value are respectively generated according to the dynamic threshold in a dynamic threshold. The noise estimation device of claim 1, wherein the noise estimation unit comprises: a noise estimator for determining the weight value corresponding to each block of the image and the The variance number is weighted and averaged to output the noise estimate. The noise estimation device of claim 8, wherein the noise estimation unit further comprises: a protection unit, wherein when the sum of the weight values of all the blocks is lower than a critical value, The noise estimation value output by the noise estimator is divided by a low value as the noise estimation value output, wherein the sum of the weight values of all the blocks is lower than the threshold value , the lower the value of the lower value. The noise estimation device of claim 1, wherein the image has M pixels, each pixel system corresponding to a block, and the target pixel data is a central pixel of the corresponding block. data. A noise estimation method for calculating a noise estimation value of an image, the image having M blocks, the M blocks may overlap or not overlap, and each block has a plurality of pixel data, The pixel data includes a target pixel data, and the M system is a positive integer greater than 1. The noise estimation method includes: according to the pixel data of the i-th block of the image, and a first a plurality of previous pixel data of the i-th block of the pre-image, and calculating a number of pixels corresponding to the plurality of pixel values to generate a pixel distribution, i is a positive integer between 1 and M; combining the images The pixel data of the ith block and the previous pixel data of the ith block of the previous image, and correspondingly generating a variogram; the ith block of the image The target pixel data is used as a reference, a distribution curve is generated according to the variance, and the pixel distribution and the distribution curve are compared to correspondingly generate a weight value; and the weight value corresponding to each block of the image and the weight The number of variances, the noise estimate is output. The method for estimating the noise according to claim 11, wherein the step of generating the pixel distribution comprises: calculating a correspondence between the pixel data of the ith block of the image to each of the pixel values a number of pixels to output a first distribution; calculating a number of pixels of the previous pixel data of the ith block of the previous image corresponding to each of the pixel values to output a second distribution; and combining The first distribution and the second distribution are to generate the pixel distribution. The method for estimating noise according to claim 11, wherein the step of generating the variation comprises: combining the pixel data of the i-th block of the image with the ith of the previous image The previous pixel data of the block is correspondingly outputted to a synthesized pixel data; and the synthesized pixel data is used to generate the variance. The method for estimating a noise according to claim 11, wherein the step of generating the weight value comprises: using the target pixel data of the i-th block of the image as a reference, according to the variation Looking up the table to output the distribution curve; comparing the distribution curve and the pixel distribution to output a difference value; and correspondingly generating the weight value according to the difference value. The noise estimation method according to claim 14, wherein the distribution curve is a Gaussian distribution. The method for estimating noise according to claim 14, wherein when the magnitude of the difference is between a first threshold and a second threshold, the second threshold is greater than the first valve. A value, when the difference value is larger, the corresponding weight value is smaller. The noise estimation method of claim 16, wherein the first threshold value and the second threshold value are respectively generated in a dynamic threshold according to the variation value. The method for estimating noise according to claim 11, wherein the step of outputting the noise estimate comprises: weighting the weight according to the weight value corresponding to each block of the image and the variance Processing to output the noise estimate. The method for estimating noise according to claim 18, wherein the step of outputting the noise estimation value further comprises: when the sum of the weight values of all the blocks is lower than a critical value, The noise estimation value output by the noise estimator is divided by a low value as the noise estimation value output, wherein the total weight value of all the blocks is increased. Below this threshold, the lower the value is greater. The method for estimating noise according to claim 11, wherein the image has M pixels, each pixel system corresponding to a block, and the target pixel data is a central pixel of the corresponding block. data.